In this project, we tried to make mechine to answer queastions after watching movie. This is a very difficult task, and the dataset is published from the link below: https://github.com/makarandtapaswi/MovieQA_benchmark
We have done many work on this task including applying different models and changing different parameters.
And so far, this model turns out to have the best performance with accuracy of 36.4% on valdation set and 56.2% on training set.
It looks like that the accuracy of training set tends to keep going up as epoches increase. Howevwer, validation set reach the peak of 36.4% and starts to decline.
